1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/ipc/namespace.c 4 * Copyright (C) 2006 Pavel Emelyanov <xemul@openvz.org> OpenVZ, SWsoft Inc. 5 */ 6 7 #include <linux/ipc.h> 8 #include <linux/msg.h> 9 #include <linux/ipc_namespace.h> 10 #include <linux/rcupdate.h> 11 #include <linux/nsproxy.h> 12 #include <linux/slab.h> 13 #include <linux/cred.h> 14 #include <linux/fs.h> 15 #include <linux/mount.h> 16 #include <linux/user_namespace.h> 17 #include <linux/proc_ns.h> 18 #include <linux/sched/task.h> 19 20 #include "util.h" 21 22 static struct ucounts *inc_ipc_namespaces(struct user_namespace *ns) 23 { 24 return inc_ucount(ns, current_euid(), UCOUNT_IPC_NAMESPACES); 25 } 26 27 static void dec_ipc_namespaces(struct ucounts *ucounts) 28 { 29 dec_ucount(ucounts, UCOUNT_IPC_NAMESPACES); 30 } 31 32 static struct ipc_namespace *create_ipc_ns(struct user_namespace *user_ns, 33 struct ipc_namespace *old_ns) 34 { 35 struct ipc_namespace *ns; 36 struct ucounts *ucounts; 37 int err; 38 39 err = -ENOSPC; 40 ucounts = inc_ipc_namespaces(user_ns); 41 if (!ucounts) 42 goto fail; 43 44 err = -ENOMEM; 45 ns = kzalloc(sizeof(struct ipc_namespace), GFP_KERNEL_ACCOUNT); 46 if (ns == NULL) 47 goto fail_dec; 48 49 err = ns_alloc_inum(&ns->ns); 50 if (err) 51 goto fail_free; 52 ns->ns.ops = &ipcns_operations; 53 54 refcount_set(&ns->ns.count, 1); 55 ns->user_ns = get_user_ns(user_ns); 56 ns->ucounts = ucounts; 57 58 err = mq_init_ns(ns); 59 if (err) 60 goto fail_put; 61 62 err = -ENOMEM; 63 if (!setup_mq_sysctls(ns)) 64 goto fail_put; 65 66 if (!setup_ipc_sysctls(ns)) 67 goto fail_mq; 68 69 err = msg_init_ns(ns); 70 if (err) 71 goto fail_put; 72 73 sem_init_ns(ns); 74 shm_init_ns(ns); 75 76 return ns; 77 78 fail_mq: 79 retire_mq_sysctls(ns); 80 81 fail_put: 82 put_user_ns(ns->user_ns); 83 ns_free_inum(&ns->ns); 84 fail_free: 85 kfree(ns); 86 fail_dec: 87 dec_ipc_namespaces(ucounts); 88 fail: 89 return ERR_PTR(err); 90 } 91 92 struct ipc_namespace *copy_ipcs(unsigned long flags, 93 struct user_namespace *user_ns, struct ipc_namespace *ns) 94 { 95 if (!(flags & CLONE_NEWIPC)) 96 return get_ipc_ns(ns); 97 return create_ipc_ns(user_ns, ns); 98 } 99 100 /* 101 * free_ipcs - free all ipcs of one type 102 * @ns: the namespace to remove the ipcs from 103 * @ids: the table of ipcs to free 104 * @free: the function called to free each individual ipc 105 * 106 * Called for each kind of ipc when an ipc_namespace exits. 107 */ 108 void free_ipcs(struct ipc_namespace *ns, struct ipc_ids *ids, 109 void (*free)(struct ipc_namespace *, struct kern_ipc_perm *)) 110 { 111 struct kern_ipc_perm *perm; 112 int next_id; 113 int total, in_use; 114 115 down_write(&ids->rwsem); 116 117 in_use = ids->in_use; 118 119 for (total = 0, next_id = 0; total < in_use; next_id++) { 120 perm = idr_find(&ids->ipcs_idr, next_id); 121 if (perm == NULL) 122 continue; 123 rcu_read_lock(); 124 ipc_lock_object(perm); 125 free(ns, perm); 126 total++; 127 } 128 up_write(&ids->rwsem); 129 } 130 131 static void free_ipc_ns(struct ipc_namespace *ns) 132 { 133 /* mq_put_mnt() waits for a grace period as kern_unmount() 134 * uses synchronize_rcu(). 135 */ 136 mq_put_mnt(ns); 137 sem_exit_ns(ns); 138 msg_exit_ns(ns); 139 shm_exit_ns(ns); 140 141 retire_mq_sysctls(ns); 142 retire_ipc_sysctls(ns); 143 144 dec_ipc_namespaces(ns->ucounts); 145 put_user_ns(ns->user_ns); 146 ns_free_inum(&ns->ns); 147 kfree(ns); 148 } 149 150 static LLIST_HEAD(free_ipc_list); 151 static void free_ipc(struct work_struct *unused) 152 { 153 struct llist_node *node = llist_del_all(&free_ipc_list); 154 struct ipc_namespace *n, *t; 155 156 llist_for_each_entry_safe(n, t, node, mnt_llist) 157 free_ipc_ns(n); 158 } 159 160 /* 161 * The work queue is used to avoid the cost of synchronize_rcu in kern_unmount. 162 */ 163 static DECLARE_WORK(free_ipc_work, free_ipc); 164 165 /* 166 * put_ipc_ns - drop a reference to an ipc namespace. 167 * @ns: the namespace to put 168 * 169 * If this is the last task in the namespace exiting, and 170 * it is dropping the refcount to 0, then it can race with 171 * a task in another ipc namespace but in a mounts namespace 172 * which has this ipcns's mqueuefs mounted, doing some action 173 * with one of the mqueuefs files. That can raise the refcount. 174 * So dropping the refcount, and raising the refcount when 175 * accessing it through the VFS, are protected with mq_lock. 176 * 177 * (Clearly, a task raising the refcount on its own ipc_ns 178 * needn't take mq_lock since it can't race with the last task 179 * in the ipcns exiting). 180 */ 181 void put_ipc_ns(struct ipc_namespace *ns) 182 { 183 if (refcount_dec_and_lock(&ns->ns.count, &mq_lock)) { 184 mq_clear_sbinfo(ns); 185 spin_unlock(&mq_lock); 186 187 if (llist_add(&ns->mnt_llist, &free_ipc_list)) 188 schedule_work(&free_ipc_work); 189 } 190 } 191 192 static inline struct ipc_namespace *to_ipc_ns(struct ns_common *ns) 193 { 194 return container_of(ns, struct ipc_namespace, ns); 195 } 196 197 static struct ns_common *ipcns_get(struct task_struct *task) 198 { 199 struct ipc_namespace *ns = NULL; 200 struct nsproxy *nsproxy; 201 202 task_lock(task); 203 nsproxy = task->nsproxy; 204 if (nsproxy) 205 ns = get_ipc_ns(nsproxy->ipc_ns); 206 task_unlock(task); 207 208 return ns ? &ns->ns : NULL; 209 } 210 211 static void ipcns_put(struct ns_common *ns) 212 { 213 return put_ipc_ns(to_ipc_ns(ns)); 214 } 215 216 static int ipcns_install(struct nsset *nsset, struct ns_common *new) 217 { 218 struct nsproxy *nsproxy = nsset->nsproxy; 219 struct ipc_namespace *ns = to_ipc_ns(new); 220 if (!ns_capable(ns->user_ns, CAP_SYS_ADMIN) || 221 !ns_capable(nsset->cred->user_ns, CAP_SYS_ADMIN)) 222 return -EPERM; 223 224 put_ipc_ns(nsproxy->ipc_ns); 225 nsproxy->ipc_ns = get_ipc_ns(ns); 226 return 0; 227 } 228 229 static struct user_namespace *ipcns_owner(struct ns_common *ns) 230 { 231 return to_ipc_ns(ns)->user_ns; 232 } 233 234 const struct proc_ns_operations ipcns_operations = { 235 .name = "ipc", 236 .type = CLONE_NEWIPC, 237 .get = ipcns_get, 238 .put = ipcns_put, 239 .install = ipcns_install, 240 .owner = ipcns_owner, 241 }; 242